@research-copilot/mcp-pdf

MCP server for PDF text and metadata extraction using unpdf.

Installation

npx @research-copilot/mcp-pdf

Or via MCP configuration:

{
  "mcpServers": {
    "research-pdf": {
      "command": "npx",
      "args": ["-y", "@research-copilot/mcp-pdf"]
    }
  }
}

Tools

pdf_extract_text

Extract text content from a PDF file.

Inputs:

• path (string, required) - Absolute path to PDF file
• mergePages (boolean, optional) - Merge all pages into single text block (default: true)

Returns:

• text (string) - Extracted text content
• totalPages (number) - Total page count

Note: unpdf extracts text in content-stream order. For two-column academic papers, text from both columns may be interleaved. Page numbers are advisory.

pdf_extract_metadata

Extract PDF metadata (page count, document info).

Inputs:

• path (string, required) - Absolute path to PDF file

Returns:

• totalPages (number) - Total page count
• info (object) - Document metadata (title, author, creation date, etc.)

Implementation

• Built with unpdf for PDF parsing
• Uses Mozilla PDF.js under the hood
• Text extraction is coarse (content-stream order, not reading order)
• Suitable for simple text extraction; complex layouts may need specialized tools

License

MIT